1 Backgrounds

Americans have been concerned about their children’s conditions in health, safety, educational progress, and moral development ever since the 1970s (K. C. Land et al. 2007, pp. 105–6) and this concern led to a sustained interest in indicators of child well-being. In addition, our interest in constructing indicators is also attributed to “a movement toward accountability-based public policy, which demands more accurate measures of the conditions children face” (Ben-Arieh 2008, p. 5). However, the quality of life (or well-being)Footnote 1 is “notoriously vague in content, and inadequate appreciation of this fact has led to a number of problems in research on the subject” (Gerson 1976, p. 794). Theoretically, both descriptive (or objective) and evaluative (or subjective) measures should be adopted to properly embrace the quality of life (Sirgy et al. 2006, pp. 346–7; Cummins 2000, p. 56).

Different researchers (e.g., Kenneth C. Land et al. 2001; Lee et al. 2009; Moore et al. 2008; Vandivere and McPhee 2008) and institutes (e.g., the Annie E. Casey Foundation; the Foundation for Child Development) have measured child well-being by adopting different well-being domains and indexing methods. For instance, the Foundation for Child Development (FCD) adopted 28 indicators in seven domains and constructed the FCD-Land Index of child well-being (Kenneth C. Land et al. 2001), whereas the Annie E. Casey Foundation selected 10 indicators without designating specific domains of child well-being to produce the Kids Count Index (The Annie E. Casey Foundation 2010).

Summary indices have been developed due to an increased supply of information about child well-being (Ben-Arieh 2008, p. 12). While having a single number would make easier for the public to know about the conditions and to hold policy makers accountable, creating an overall index can be a challenge for researchers (Moore et al. 2007, p. 292). As Moore and colleagues (2008) pointed out, different trends of individual indicators may be masked once they are included in an overall index (p. 19) and “inevitably, the indicators used to capture the domain constructs are constrained by available data” (p. 20). One of the major criticisms about constructing composite indices is that essential components can be excluded when making a single index (see Booysen 2002, for the summary of criticisms).

Sustained efforts have been made to compare child well-being among different countries (e.g., T. E. Jordan 1983; Thomas E. Jordan 1993; Bradshaw and Richardson 2009) or among provinces within a nation (e.g., Kenneth C. Land et al. 2001; Casas et al. 2007; Bradshaw et al. 2009; Hanafin and Brooks 2009; Vandivere and McPhee 2008). For instance, Jordan (1983; 1993) compared child well-being among 122 countries using a national index of children’s quality of life score, consisting of nine variables largely from the UNICEF reports on the conditions of world children. In Bradshaw and Richardson’s (2009) study, 29 European countries were compared and ranked based on a single composite index as well as seven domain indices in child well-being such as health, subjective well-being, children’s relationship, material situation, risk and safety, education, and housing and environment.

While a national level or a state level comparison of child well-being has been conducted by many studies, only a few studies (e.g., Menanteau-Horta and Yigzaw 2002; Lee et al. 2009; Niclasen and Kohler 2009) investigated child well-being at a local level. For instance, Lee and others (2009) constructed five domain-specific indices of child well-being for each county in the San Francisco Bay Area, aggregated them into an equally weighted composite child well-being index (CWI), and tracked the changes of overall CWI over a decade. Menanteau-Horta and Yigzaw (2002) compared each of 16 child welfare indicators with a composite index of social well-being at a county level in Minnesota, focusing on the comparison between rural and metropolitan counties.

We believe child well-being research at a local level brings practical benefits to citizens paricularly with children. In his seminal article “A Pure Theory of Local Expenditures,” Charles Tiebout (1956) argues that citizens may choose the community that best satisfies their preference for public goods (p. 418) unlike the argument of classical theories in public finance (e.g., Musgrave 1939; Samuelson 1954). Although the Tiebout’s model has limitations in its assumptions such as citizen’s full mobility and full knowledge of differences among communities (see p. 419 for all six assumptions), it can be applied to the case of child well-being, too. Given the growing concerns about child well-being among parents, the level of child well-being at local communities can be one of the critical factors that would affect the decision about where to live. Different studies examined a possibility that child well-being can be affected by neighborhood characteristics. For instance, neighborhood characteristics have influence on child maltreatment (Coulton et al. 2007; McDonell and Skosireva 2009), child health (Lumeng et al. 2006; Xue et al. 2007), child safety (Lumeng et al. 2006; O’Campo et al. 2000; McDonell and Skosireva 2009), and education outcomes (McWayne et al. 2007; Ceballo et al. 2004).

In the present study, we investigated child well-being in North Carolina counties because counties, as administrative units between states and municipalities in the U.S., largely determine the living conditions for their residents through delivering social services to their residents (Menanteau-Horta and Yigzaw 2002, p. 711). To measure child well-being, we chose four domains such as health, safety, education, and economic well-being, each of which consisted of four indicators (see Table 1 for their definitions and sources). For each domain, we constructed a composite index with equal weights to component indicators. An overall index was also constructed, based on composite indices of all domains with equal weights to component domains. In the end, counties were ranked based on each domain’s index as well as an overall index. In addition, child well-being was also compared between urban and rural areas and among four physiographic regions such as Blue Ridge, Piedmont, Inner Coastal, and Outer Coastal regions. The findings in the present study would not only help citizens make a more informed decision about where to live and where to raise their children, but also provide policy makers and implementers an idea about the strengths and weaknesses in their communities and what they should do to make their communities more attractive.

Table 1 Indicators and data sources
Full size table

2 North Carolina

North Carolina is the fourth fastest-growing state in the United States with over 9.3 million population and 16.6% population change rateFootnote 2 (2009 estimate, U.S. Census Bureau 2010). North Carolina has a multitude of rich resources and public goods and services for citizens in 100 counties that extend across 48,710 square miles (U.S. Census Bureau 2010). While boasting of one of the country’s best higher public educational systems, North Carolina communities are also fighting issues such as crimes and unemployment. About 4.74 violent crimes and 40.8 property crimes per 1,000 people were reported during 2008 (State Bureau of Investigation 2009) and North Carolina’s unemployment rate was 11.2% as of December 2009 (Bureau of Labor Statistics 2009).

Rapidly growing technology, shifting workforce demographics, diverse populations and an ever-shifting economy have opened the doors to allow citizens to become more mobile and migratory. Although theoretically quality of life for citizens in North Carolina should be equal regardless of the county in which they reside, communities and public officials often struggle to meet the demands of the public while facing decreasing tax revenues. North Carolina communities are indeed in a competitive market to attract and retain citizens and to generate tax revenues above a certain level. Citizens look for communities that match their specific sets of needs and have reasonable tax structures. Given the revenue and expenditure patterns, one could assume that citizens would move to the community that can best satisfy their needs (Tiebout 1956).

Citizens, and specifically children, thrive when their community provides resources and support that include adequate health care, safe neighborhoods, quality education, and stable economic well-being. If a citizen values a quality education system, they can seek out the community that will provide the best educational outcomes for their family. Another citizen may value the availability of quality healthcare, safe neighborhoods, or stable economic well-being in their community and make their community choice based upon those values and needs.

3 Methodology

Before comparing child well-being among counties, we constructed an index for each domain and an overall index with an equal weight to each indicator and to each domain, respectively. With constructed indices, a county’s child well-being was compared in three ways. First, counties were grouped into urban and rural areas, and child well-being was compared between groups (Table 2). Second, counties were grouped into four physiographic regions, and child well-being was compared among groups (Table 3). Third, counties were ranked based on their composite indices (see Table 4 for the summary of those ranks, and Appendices 4–8 for details in overall and four domains). To compare groups, we conducted t-test for comparing urban and rural counties, and analysis of variance (ANOVA) for comparing physiographic regions. According to the Kolmogorov-Smirnov test results,Footnote 3 all indices were normally distributed.

Table 2 t-test results for urban and rural county comparison
Full size table
Table 3 ANOVA test results for physiographic regions
Full size table
Table 4 County ranks
Full size table

3.1 Selection of Domains and Indicators

Cummins (1996) proposed seven comprehensive quality of life (ComQoL) domains that were believed as import aspects of lives by most people after reviewing 173 terms and 27 definitions used to describe quality of life in the literature. Land and his colleagues (2001) applied these ComQoL domains to child well-being, and suggested seven constituent domains for child well-being such as health, material well-being, educational attainment, safety/behavioral concerns, emotional/spiritual well-being, social relationships (i.e., with family and peers), and participation in schooling, which have been used as a conceptual guide for many child well-being studies (e.g., Lau and Bradshaw 2010; Bradshaw and Richardson 2009; Richardson et al. 2008; Lee et al. 2009).

In the present study, the selection of component domains for child well-being was based on Land and his colleagues’ (2001) suggestion. Although all suggested domains were not adopted due to data unavailability, we believe that major aspects of child well-being could be measured with these domains. Internal reliability among these selected domains was sufficient since the computed Cronbach’s alpha value was 0.83.Footnote 4 In other words, the selected domains measured the same construct of child well-being. Principal component analysis was also conducted to see how many factors could be possibly extracted out of four component domains. As seen in part (a) of Appendix 1, only one factor (i.e., comp1) had a larger eigenvalue than one, which justifies retaining only one factor (Kim and Mueller 1978). This conclusion was also supported in the Scree plot (see part (b) in Appendix 2) because the Scree test suggests to stop factoring at the point when eigenvalues begin to level off and form a straight line (Cattell 1965). In sum, child well-being was properly measured by the selected domains in the present study.

In each domain, we selected four indicators, most of which measured negative constructs of child well-being. Out of 16 indicators, there were only four positive indicators that measured positive constructs of child well-being such as SAT score, proficient rates for third and eighth grade students from the education domain, and median household income from the economic well-being domain. Better outcomes are not guaranteed by simply having more inputs. Therefore, we selected more outcome-related indicators in each domain as suggested in the child well-being literature (e.g., K. C. Land et al. 2007; Moore et al. 2008). Selection of indicators is basically constrained by data availability. However, selected indicators for a certain domain are supposed to measure the same construct to produce a meaningful composite index in the end. According to the computed Cronbach’s alpha values, the internal reliability of selected indicators in most domains was acceptable.Footnote 5

3.2 Composite Index Making

Indicators’ values were first transformed into standard scores (i.e., z-values) because they had different scales. In fact, this is one of the popular methods in composite indexing (see Booysen 2002, pp. 123–126 for popular four methods of scaling composite index) and has been widely adopted (e.g., Gallardo et al. 2009; Norton 2007). If positive indicators were incorporated with negative indicators without changing the signs of their standard scores, we would not know whether a higher index score means good news or bad news to communities. Therefore, when transforming positive indicators that measured positive constructs of child well-being, we changed the signs of their standard scores before making composite indices. As a result, higher index scores mean lower levels of child well-being in the present study. To construct a composite index of each domain, standard scores of component indicators were summed up and divided by the number of indicators in a domain (i.e., four in our study) with equal weights to indicators. An overall index was constructed in the same way. That is, composite indices of component domains were added up and divided by the number of domains (i.e., four in our study) with equal weights to domains.

The usefulness of a summary index has been questioned due to disagreement on different weights given to its components among individuals, policy makers, and researchers themselves (M. R. Hagerty and Land 2007, p. 457). Still different summary indices have been suggested without justifying why equal or different weights were given to their components (Michael R. Hagerty et al. 2001). According to the results of principal component analysis conducted in the present study, each component domain similarly contributed to forming a summary index (see eigenvector table in Appendix 1).Footnote 6 Statistically, these results of principal component analysis can be a rationale for giving an equal weight to component domains. Even in this case, however, individuals may have different preferences and may not agree with the idea that component domains have equal importance in child well-being. Therefore, unless we have information about average weights of importance given to each component domain among individuals, equal weighting can be a minimax estimator that will make minimal individuals’ disagreement about the importance of component domains (M. R. Hagerty and Land 2007, p. 486).

3.3 County Grouping

In addition to ranking counties (explained in Section 4.4), we grouped counties into urban and rural areas and compared child well-being between groups. Although there are diverse definitions of urban and rural areas (e.g., Belanger and Stone 2008; Landsman 2002; Whitaker 1984), we regarded metropolitan counties as urban areas, and non-metropolitan counties as rural areas. In the present study, counties with more than 50,000 residents were grouped into urban areas and counties with less than 50,000 residents were defined as rural areasFootnote 7 as a common approach (Belanger and Stone 2008, p. 103). According to this definition, North Carolina had 40 urban counties and 60 rural counties.

Counties were also grouped into four physiographic regions. North Carolina has physiographic regions such as Blue Ridge (or Mountains), Piedmont, Inner Coastal, and Outer Coastal (or Tidewater) regions (Gade et al. 2002). When a county was spanned over two physiographic regions, we assigned it to one specific physiographic region whose area seemed larger than that of the other region according to the North Carolina Geological Survey map (2004). Footnote 8 As a result, North Carolina’s counties consisted of 17 Blue Ridge, 41 Piedmont, 30 Inner Coastal, and 12 Outer Coastal regions.

4 Results

4.1 Comparing Urban and Rural Counties

Table 2 summarizes t-test results for the comparison between urban and rural counties (see Appendix 2 for details). Although urban counties had a significantly higher level of child well-being (i.e., lower index scores) than rural counties in overall and economic well-being, no considerable differences existed between urban and rural counties in other domains such as health, safety, and education. Therefore, it is obvious that the overall index was substantially influenced by a huge difference in economic well-being between urban and rural counties. Generally speaking, however, urban counties appear to have a better status (i.e., lower index scores) than rural counties in all component domains as seen in Fig. 1.

Fig. 1
figure 1

Index score comparison between urban and rural counties

Full size image

4.2 Comparing Physiographic Regions

Analysis of variance (ANOVA) test was conducted to compare child well-being among four physiographic regions. Since the health domain did not meet the equal variance requirement for ANOVA test (see Appendix 3a for details), we ran Kruskal-Wallis (K-W) test for the health domain to verify ANOVA results for health and found no difference between ANOVA and K-W test results (K-W results not shown). As seen in Table 3, physiographic regions in North Carolina were significantly different in all domains of child well-being.

By running a post-hoc test (the Sheffee method was used), we could identify which regions were better in a certain domain of child well-being, comparing with other regions. As seen in Appendix 3d, the Inner Coastal region had a significantly higher index score (i.e., a lower level of child well-being) than the other three regions in overall child well-being and most component domains except economic well-being.Footnote 9 In the economic well-being domain, only the Piedmont region had a significantly lower index score (i.e., a higher status) than the Inner Coastal region. In the health domain, the Blue Ridge region showed a significantly lower index score (i.e., a higher status) than the Piedmont region. As seen in Fig. 2 that shows a simple comparison among regions, the Blue Ridge and the Outer Coastal regions appear to have a higher level of child well-being than the Piedmont and the Inner Coastal regions in all domains except the economic well-being domain.

Fig. 2
figure 2

Index score comparison among physiographic regions

Full size image

4.3 Comparing Urban and Rural Counties in Physiographic Regions

Useful information was revealed when we compared urban and rural counties in each physiographic region. In North Carolina, the Blue Ridge and the Outer Coastal regions have much more rural counties than urban counties, whereas the Piedmont region has more urban counties than rural counties.Footnote 10 As seen in Fig. 3, urban counties appear to have a better status (i.e., a lower score) than rural counties in overall child well-being and all component domains, regardless of physiographic regions. According to the t-test results (results not shown) that compared urban and rural in each region, overall child well-being was significantly different only in the Piedmont region (p < 0.05), whereas economic well-being was significantly different in most regions such as the Blue Ridge, the Piedmont, and the Inner Coastal regions (p < 0.05). In other domains such as health, safety, and education, however, urban areas were not significantly different from rural areas, regardless of physiographic regions. It needs to be also noted that although urban and rural counties appeared to be different in the Outer Coastal region, it was not supported by t-test results because there is only one urban county in that region.

Fig. 3
figure 3

Urban vs. rural comparison in physiographic regions

Full size image

4.4 County Rankings

North Carolina (NC) counties were ranked, based on their index scores. Recall that a lower index score indicates a higher status of child well-being in the present study. The top 10 counties in overall and four component domains are summarized in Table 4 (see Appendices 4–8 for complete lists of county ranking). The relationships between overall index and component domain indices and between component domains were also examined. Since all relationships were significantly positive (p < 0.01) as seen in Table 5, a county that had a high ranking in the overall index was more likely to rank high in most component domains. For example, Watauga, the number one county in the overall index was also ranked seventh in health, ranked fifth in safety, and ranked first in education.

Table 5 Pearson product correlations
Full size table

5 Discussion

This cross-sectional study was conducted with an assumption that the level of child well-being would not change in the future, and we did not take into account county governments’ previous efforts to improve child well-being in their counties. Although it is important to understand how much effort has been made and can be made in the future to improve child well-being in county jurisdictions, current county rankings can be also useful and help citizens decide where to live. While rankings can be good information to perspective residents, it should be also noted that low ranked counties do not necessarily mean they are bad places to live. Regardless of ways to rank counties, there should be high ranked counties and low ranked counties. Theoretically, it is possible that all North Carolina counties provide better environments to children than any other places in the U.S. and vice versa. In any case, however, it is not hard to imagine that parents want to provide their children with higher quality of life by relocating to a community that can provide better environments. From an administrator’s perspective, county rankings can help them understand where their counties are located among peer counties, and motivate them to remedy certain areas that need to be improved. Through benchmarking peer governments that are advanced in a certain area, administrators in charge would make more informed decision and policy outcomes can be improved as a result (D. Ammons 1996; D. N. Ammons et al. 2001).

Disadvantages in rural areas have been investigated from different perspectives (e.g., Belanger and Stone 2008; Liao 2009; Menanteau-Horta and Yigzaw 2002), with an assumption that “whatever the subject or indicator—per capita income, health care, education, employment opportunity, or transportation—the nonmetropolitan people of the nation have less” (Ginsberg 1998, p. v). According to the findings in the present study, it is not really true in most child well-being domains. Only in economic well-being was there a huge difference between urban and rural counties, which substantially contributed to a significant difference in overall child well-being (see t-test results in Table 2 and a comparison graph in Fig. 1). Even though urban counties were not significantly different from rural counties in most child well-being domains, urban counties provided a little bit better environments for child well-being (i.e., a lower index score) than rural counties as seen in Figs. 1 and 3.

While the findings in the present study can be useful to county officials, state legislative leaders, citizens, and others, our research has limitations. First, data at a local level are very limited unlike data at a state level or at a national level. Our study is not an exception. The selection of indicators depended largely on data availability. Although survey-based data are more reliable than report-based data in which different filtering mechanisms may be involved, we had to use some report-based data (e.g., child abuse rate) because they were only available. Second, a single index has a limitation in itself because it cannot reflect the trends of individual indicators while comparison among counties can be easily conducted by using it (Moore et al. 2008). Third, changes in child well-being may be important but cannot be measured by cross-sectional studies like this research. While the present study provides a good comparison among NC counties at a certain time point, only longitudinal studies can identify a trend in which we can see how much improvement has been made.